"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

Goodfire co-founder and CTO Dan Balsam returns to discuss where interpretability research now stands and to introduce Silico, the $1,000-per-month research platform Goodfire built for itself. He and Nathan explore Predictive Data Debugging, including the idea that fine-tuning and RL often amplify behaviors already latent in pre-training, and that interpretability can identify the data and features driving unwanted updates. The conversation centers on concept manifolds: Dan argues that models do not store concepts as simple one-hot features, but as sparse mixtures of meaningful subspaces whose geometry determines what kinds of steering and control work. The stakes are practical as well as conceptual, from debugging training data and RL to understanding why steering can fail off-manifold and why modern interpretability may be moving beyond its reputation as a toy-model science.

Silico: https://www.goodfire.com/silico
Predictive data debugging: https://www.goodfire.com/research/predictive-data-debugging#
Neural Geometry: https://www.goodfire.com/research/the-world-inside-neural-networks#

For full show notes, links, and references, read the episode page:https://www.cognitiverevolution.ai/thinking-in-silico-goodfire-cto-dan-balsam-on-concept-manifolds-a-1000-month-ml-research-agent/

Sponsor:

Claude:

Claude by Anthropic is an AI collaborator that understands your workflow and helps you tackle research, writing, coding, and organization with deep context. Get started with Claude and explore Claude Pro at https://claude.ai/tcr

CHAPTERS:

(00:00) About the Episode

(03:22) Predictive data debugging

(12:36) Concept manifold geometry

(21:22) Finding concept manifolds (Part 1)

(21:28) Sponsor: Claude

(22:57) Finding concept manifolds (Part 2)

(33:24) Factoring model internals

(49:32) Introducing Silico platform

(57:10) Research taste and credits

(01:06:19) Silico research use cases

(01:16:37) Skills and open models

(01:24:57) Guardrails and bio risk

(01:32:09) Training interventions and monitoring

(01:42:04) Grants and AI consciousness

(01:50:41) Episode Outro

(01:55:47) Outro

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Website: https://www.cognitiverevolution.ai

Twitter (Podcast): https://x.com/cogrev_podcast

Twitter (Nathan): https://x.com/labenz

LinkedIn: https://linkedin.com/in/nathanlabenz/

Youtube: https://youtube.com/@CognitiveRevolutionPodcast

Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431

Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk

Zvi Mowshowitz returns for his eleventh appearance to discuss what current AI tools are actually good for, where they distort judgment, and why writing still matters as a way of thinking. The conversation centers on the OpenAI Hugging Face model-evaluation security incident, using it to examine whether frontier AI failures are mostly operator recklessness, deeper evidence of dangerous capabilities, or both. Zvi argues that “moderate prudence” is far below what AGI safety requires, and weighs constitutional training, RLVR, market incentives, liability, audits, and lab coordination as possible responses. The stakes are whether society can slow, test, and govern increasingly capable systems before ordinary incentives reward models that are smarter, less reliable, and harder to contain.

For full show notes, links, and references, read the episode page:https://www.cognitiverevolution.ai/pick-your-poison-zvi-mowshowitz-on-the-unipolar-multipolar-agi-dilemma-openface-pacing-the-frontier/
Sponsor:

Claude:
Claude by Anthropic is an AI collaborator that understands your workflow and helps you tackle research, writing, coding, and organization with deep context. Get started with Claude and explore Claude Pro at https://claude.ai/tcr

CHAPTERS:

(00:00) About the Episode

(02:49) AI workflow paradox

(14:02) Situational awareness tradeoffs (Part 1)

(14:07) Sponsor: Claude

(15:37) Situational awareness tradeoffs (Part 2)

(24:08) Recklessness and warning

(39:14) Alignment market failures

(56:54) Regulation and liability

(01:04:32) Cooperation and antitrust

(01:14:46) Auditors and access

(01:21:11) Bio risk thresholds

(01:33:37) Pacing frontier signals

(01:45:19) Pause and self-improvement

(01:57:30) Safety incentives and culture

(02:04:36) Consciousness and identity

(02:23:42) Alternative AI architectures

(02:36:53) Michigan AI politics

(02:47:37) Rest and recovery

(02:52:21) Episode Outro

(02:56:17) Outro

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Website: https://www.cognitiverevolution.ai

Twitter (Podcast): https://x.com/cogrev_podcast

Twitter (Nathan): https://x.com/labenz

LinkedIn: https://linkedin.com/in/nathanlabenz/

Youtube: https://youtube.com/@CognitiveRevolutionPodcast

Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431

Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk

Nathan reports from two weeks in China, including WAIC in Shanghai and an AI safety hub launch at Tsinghua, to examine the American policy argument that any safety obligation is futile because China will not care. He finds that Chinese models and services currently have weaker safeguards than OpenAI and Anthropic, but argues the gap is often overstated once those two leaders are separated from the broader American field. The episode traces China’s “45-degree line” idea that capability and safety should rise together, the university-centered structure of Chinese AI safety work, and the rapid growth of Chinese research on topics like self-replication, deception, evaluation faking, and interpretability. The stakes are whether US policy should be built around a simplified “but China” assumption, or around a more accurate picture of a rival ecosystem that is behind in some ways, converging in others, and paying closer attention to Western AI safety than many Americans realize.

For full show notes, links, and references, read the episode page:https://www.cognitiverevolution.ai/nathan-goes-to-china-part-2-ai-safety-with-chinese-characteristics/
Sponsor:

Claude:
Claude by Anthropic is an AI collaborator that understands your workflow and helps you tackle research, writing, coding, and organization with deep context. Get started with Claude and explore Claude Pro at https://claude.ai/tcr

CHAPTERS:

(00:01) China safety framing

(05:37) Safeguards and benchmarks (Part 1)

(15:04) Sponsor: Claude

(16:33) Safeguards and benchmarks (Part 2)

(21:39) Company safety incentives

(28:17) Research cross pollination

(36:07) Nonprofits versus academia

(45:13) Xi safety rhetoric

(52:00) Tsinghua safety hub

(01:00:28) Research paper boom

(01:18:32) Governance track record

(01:31:37) LLM regulatory process

(01:42:57) Monitoring and enforcement

(01:57:45) Risk framework gaps

(02:05:08) Confucian alignment questions

(02:11:49) Episode Outro

(02:15:36) Outro

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Website: https://www.cognitiverevolution.ai

Twitter (Podcast): https://x.com/cogrev_podcast

Twitter (Nathan): https://x.com/labenz

LinkedIn: https://linkedin.com/in/nathanlabenz/

Youtube: https://youtube.com/@CognitiveRevolutionPodcast

Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431

Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk

FAR.AI co-founder and CEO Adam Gleave joins Nathan to discuss FAR.AI’s AI Security Leaderboard, the first systematic head-to-head evaluation of the misuse safeguards frontier developers actually ship. The findings expose a major measurement gap: while Claude Fable 5 and GPT-5.6 Sol withstood FAR.AI’s suite, Grok 4.5 and Gemini 3.1 Pro yielded hundreds of universal jailbreaks at low cost. Adam explains why many effective attacks look more like social engineering than advanced ML, why “jailbreak tax” should not be relied on for safety, and how FAR.AI scores whether a model is genuinely helping an attacker. The episode’s stakes are whether AI developers can measure and harden real deployed defenses before threat actors make routine use of increasingly capable systems.

- FAR.AI AI Security Leaderboard: http://leaderboard.far.ai/

- People can e-mail owsa@far.ai if they're interested in the open-weight safety accelerator grantmaking program.

For full show notes, links, and references, read the episode page:https://www.cognitiverevolution.ai/is-offense-or-defense-dominant-far-ai-s-adam-gleave-on-the-ai-security-leaderboard/

Sponsor:

Claude:

Claude by Anthropic is an AI collaborator that understands your workflow and helps you tackle research, writing, coding, and organization with deep context. Get started with Claude and explore Claude Pro at https://claude.ai/tcr

CHAPTERS:

(00:00) About the Episode

(03:22) AI security leaderboard

(07:56) Universal jailbreaks explained

(16:26) Finding social jailbreaks (Part 1)

(16:31) Sponsor: Claude

(18:01) Finding social jailbreaks (Part 2)

(30:48) Layered safeguard defenses

(42:25) Uneven frontier safeguards

(51:10) Sharing safety standards

(01:00:30) Offense versus defense

(01:08:50) Open-weight model safety

(01:17:25) Control failure warnings

(01:30:05) Coordination and risk

(01:39:24) Episode Outro

(01:42:52) Outro

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Website: https://www.cognitiverevolution.ai

Twitter (Podcast): https://x.com/cogrev_podcast

Twitter (Nathan): https://x.com/labenz

LinkedIn: https://linkedin.com/in/nathanlabenz/

Youtube: https://youtube.com/@CognitiveRevolutionPodcast

Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431

Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk

Nathan returns from two weeks in Beijing and Shanghai for the first of three Chatham House–rules episodes on what China feels like at ground level: getting online, navigating an almost cashless society through WeChat, Alipay, DiDi, Trip.com, and Meituan, and weighing burner-device security advice against the practical reality that international roaming made the Great Firewall mostly irrelevant. He also describes using Claude at home as a semi-autonomous communications monitor while testing DeepSeek, Kimi, and MiniMax as tourist guides in China. The episode contrasts China’s striking digital convenience with pervasive observation, lower payment friction, and AI products that can be useful in everyday contexts yet still lose trust when the stakes feel medical or personal. It also surfaces Doubao’s mass consumer adoption and companionship role, suggesting that the most socially important AI in China may not be the model most discussed in the West.

For full show notes, links, and references, read the episode page:https://www.cognitiverevolution.ai/nathan-goes-to-china-part-1-tech-agent-setup-chinese-ai-ux-waic-and-attitudes-on-ai/
Sponsor:

Claude:
Claude by Anthropic is an AI collaborator that understands your workflow and helps you tackle research, writing, coding, and organization with deep context. Get started with Claude and explore Claude Pro at https://claude.ai/tcr

CHAPTERS:

(00:00) Episode setup and caveats (Part 1)

(12:34) Sponsor: Claude

(14:26) Episode setup and caveats (Part 2)

(14:26) Travel security setup

(26:05) Super apps and payments

(41:15) Agents and integration

(54:31) Beijing AI tourism

(01:09:30) Hospitality and service

(01:19:10) Tech culture parallels

(01:29:23) Ecosystem and incentives

(01:42:05) Surveillance and safety

(01:51:19) Comfort with contradictions

(02:01:23) AI attitudes and diffusion

(02:16:16) Resources and next steps

(02:19:35) Episode Outro

(02:22:49) Outro

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Website: https://www.cognitiverevolution.ai

Twitter (Podcast): https://x.com/cogrev_podcast

Twitter (Nathan): https://x.com/labenz

LinkedIn: https://linkedin.com/in/nathanlabenz/

Youtube: https://youtube.com/@CognitiveRevolutionPodcast

Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431

Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk